The render target should be a 1 x 1 texture with 3 floating-point channels (e.g. HalfVector4). The luminance info is stored in this order: (minimum, average, maximum). That means, the minimum luminance is stored in the "red" channel, and so on.

This post-processor stores luminance information of a frame for use in the next frame. The information is stored with the currently active camera node. When there is a cut in the scene (i.e. a new level is loaded or the view changes significantly), the method InvalidateViewDependentData of the camera node needs to be called to reset the luminance information. Further, this post-processor expects that it is called once per frame for a certain camera node. It might not work as expected if it is called several times per frame (e.g. to process different views using the same camera).

Reference